Time-Sensitive Networking (TSN) is a set of IEEE 802.1 standards that make standard Ethernet deterministic. It gives every device a shared sense of time, schedules time-critical traffic into reserved windows, lets urgent frames interrupt long ones, and can send critical streams over two paths at once. The result is bounded latency, low jitter and high reliability, so control traffic and ordinary data can share one network in cars, factories and robots.
The short video above introduces TSN. This post explains what problem it solves, which standards do what, how the scheduling works with numbers, and where it is used.
Key takeaways
- TSN is a toolbox of IEEE 802.1 standards, not a single protocol. Systems pick the tools they need, usually through a published profile.
- IEEE 802.1AS (gPTP) gives all nodes a common clock, typically aligned to well under a microsecond.
- IEEE 802.1Qbv (time-aware shaper) opens and closes transmit gates on a repeating schedule, so critical traffic never waits behind bulk traffic.
- Frame preemption (IEEE 802.1Qbu with IEEE 802.3br) lets an urgent frame interrupt a long one, cutting worst-case blocking from about 12 us to about 1 us at 1 Gb/s.
- IEEE 802.1CB sends duplicate frames over separate paths and removes the copies at the far end, for seamless redundancy.
Why ordinary Ethernet is not deterministic
Standard switched Ethernet is "best effort". A frame entering a switch waits in a queue behind whatever is already there. Priorities (IEEE 802.1Q traffic classes) help, but a high-priority frame can still wait for a full-size low-priority frame that has already started transmitting, and queue depths vary with load. Latency is usually low but not bounded, and jitter depends on what everyone else is sending.
That is fine for web pages and file copies. It is not fine for a motor controller that must receive a set-point every 250 us, a brake-by-wire message, or a camera stream feeding an emergency braking function. Before TSN, those systems used dedicated fieldbuses (CAN, FlexRay, PROFIBUS, EtherCAT and others) alongside Ethernet. TSN lets one Ethernet network carry both.
Where TSN came from
TSN grew out of Audio Video Bridging (AVB), a set of IEEE standards developed in the late 2000s to stream professional audio and video over Ethernet with guaranteed bandwidth and latency. In 2012 the IEEE AVB Task Group was renamed the Time-Sensitive Networking Task Group and broadened its scope to industrial, automotive and other control applications. Many of the individual amendments have since been rolled into the base IEEE 802.1Q bridging standard.
The main TSN standards
| Function | Standard | What it does |
|---|---|---|
| Time synchronisation | IEEE 802.1AS (gPTP) | A profile of IEEE 1588 PTP that aligns every node's clock to a grandmaster |
| Scheduled traffic | IEEE 802.1Qbv | Time-aware shaper: per-queue gates opened and closed by a cyclic gate control list |
| Frame preemption | IEEE 802.1Qbu and IEEE 802.3br | Express frames interrupt preemptable frames, which resume afterwards |
| Credit-based shaping | IEEE 802.1Qav (from AVB) | Smooths audio and video streams to their reserved bandwidth |
| Redundancy | IEEE 802.1CB (FRER) | Frame replication and elimination for reliability over disjoint paths |
| Policing | IEEE 802.1Qci (PSFP) | Per-stream filtering and policing, so a faulty node cannot flood critical queues |
| Other shapers | IEEE 802.1Qch (CQF), IEEE 802.1Qcr (ATS) | Cyclic queuing and forwarding; asynchronous traffic shaping without a shared schedule |
| Configuration | IEEE 802.1Qcc | Centralised or distributed models for reserving streams and computing schedules |
Profiles then say which of these tools a given industry must use: IEC/IEEE 60802 for industrial automation, IEEE 802.1DG for automotive in-vehicle networks, and IEEE 802.1CM for mobile fronthaul.
How TSN scheduling works
1. Everyone agrees on the time
With IEEE 802.1AS, one node acts as grandmaster. Every link measures its own propagation delay with peer-delay messages, and each bridge corrects for its residence time, so all clocks in the network line up closely. Without this shared time, a schedule would mean nothing.
2. Gates open on a schedule
Each egress port has up to eight queues. With IEEE 802.1Qbv, each queue has a gate, and a gate control list says which gates are open in each slice of a repeating cycle. A typical schedule reserves a short window at the start of each cycle for control traffic only, then opens the gates for everything else.
3. Long frames must not overrun the window
If a 1,500-byte frame starts just before the control window, it delays the window. TSN handles this either with a guard band (no new low-priority frame may start if it cannot finish before the window) or with frame preemption, which pauses the long frame and resumes it later.
Worked example: a 1 ms cycle on a 1 Gb/s link
The numbers below are illustrative, using standard Ethernet overheads (8-byte preamble and 12-byte inter-frame gap).
- A 128-byte control frame takes 128 + 20 = 148 byte times on the wire. At 1 Gb/s, one byte takes 8 ns, so the frame takes 1.18 us.
- A 50 us control window therefore fits about 42 such frames per cycle.
- A maximum-size VLAN-tagged frame of 1,522 bytes takes 1,542 byte times, or 12.3 us. Without preemption, a guard band of that length is needed before each control window, which wastes about 1.2% of a 1 ms cycle per window.
- With preemption, a long frame can be paused once it has sent at least a minimum fragment, so the worst-case wait drops to roughly 120 to 160 byte times, around 1 to 1.3 us at 1 Gb/s.
- On a 100 Mb/s automotive link (100BASE-T1), everything is ten times slower: a full-size frame blocks for about 123 us, which is why preemption matters even more there.
Where TSN is used
- Automotive: zonal and central-compute vehicle architectures replace many dedicated buses with an Ethernet backbone. TSN carries camera, radar and control traffic with guaranteed latency alongside infotainment and software updates.
- Industrial automation: motion control, PLC-to-drive communication and converged IT and OT networks, often with OPC UA over TSN.
- Robotics: tight synchronisation of joints, sensors and controllers across a single network.
- Professional audio and video: the original AVB use case, still widely deployed.
- Mobile networks: 5G fronthaul, where IEEE 802.1CM defines how TSN bridges carry radio traffic.
What TSN means for chip and system designers
- Hardware timestamping in the MAC or PHY is essential; software timestamps add too much jitter.
- Per-queue gate logic and a gate control list engine are needed in every TSN-capable switch port and many endpoints.
- Preemption requires changes in the MAC (the IEEE 802.3br MAC merge sublayer), not just in software.
- Stream tables for policing and replication add on-chip memory and lookup logic.
- Verification must cover time: schedule correctness across clock adjustments, preemption at every legal boundary, and behaviour when a node loses synchronisation.
Common misconceptions
- "TSN makes Ethernet faster." It makes latency bounded and predictable. Average latency may even rise slightly when gates are closed.
- "TSN is one protocol." It is a family of standards; two TSN devices only interoperate if they support the same profile and features.
- "Any switch can run TSN in software." Time-aware scheduling and preemption need hardware support in the switch and end stations.
FAQ
Is TSN the same as AVB?
AVB is the predecessor. TSN includes the AVB standards and adds scheduling, preemption, redundancy and policing for control applications.
Does TSN need a special cable or PHY?
No. It runs over standard Ethernet physical layers, including automotive single-pair PHYs such as 100BASE-T1 and 1000BASE-T1. The PHY should support accurate timestamping.
How accurate is TSN time synchronisation?
IEEE 802.1AS is designed for sub-microsecond accuracy across several hops; well-designed systems achieve far better in practice.
Who configures the schedule?
In the fully centralised model of IEEE 802.1Qcc, a central network configuration entity computes the gate schedules and pushes them to the bridges, often from requests made by a central user configuration entity on behalf of the applications.
Your comments will be moderated before it appears here.